Velociraptor 600GB, worth the money

[VirtualRain]

There's no debating the 1E4 spec on write cycles. I think what's up for debate is what the real-world impact of this is... a few points...

The whole argument that this is unproven tech seems uninformed to me. Nothing about NAND flash is new or rocket science. NAND flash storage has been around since 1995. What's recent is that it's gotten affordable enough to assemble in parallel arrays for mass storage devices. The controller tech to manage the parallel NAND channels is what's new and where some of the initial issues were discovered... which were primarily the result of vendors attempting to rush products to market by using off-the-shelf controllers that weren't specifically designed for the task. However, all current drives run controllers that have been specifically designed or redesigned for managing parallel NAND flash arrays. So I fail to see what is unproven. NAND is well proven technology, and the controllers have been proven to address former issues, so there's really nothing left to prove.

Getting back to facts... a major factor in SSD longevity is write amplification. Intel went to great lengths to reduce write amplification to 1.1, even with it's first gen drives, which is remarkable and with wear leveling, means that the drive will last for 5 years with 100GB of writes per day. (source).

That's extreme write conditions by any measure, and probably 10 to 100 times the average writes done by a typical desktop user. Hence for a typical desktop user, an SSD could easily last decades.

In addition, Anand says this about the Intel/Micron NAND flash...

Intel and Micron have four joint fabs manufactured under the IMFT partnership, and these are the fabs that produce the flash going into Intel's SSDs. The 50nm flash used in the launch drives are rated at 10,000 erase/programming but like many of Intel's products there's a lot of built in margin. Apparently it shouldn't be unexpected to see 2, 3 or 4x the rated lifespan out of these things, depending on temperature and usage model obviously.

People should stick to facts and accurate real-world interpretation of them.

 

[slughead]

NAND is well proven technology, and the controllers have been proven to address former issues, so there's really nothing left to prove.

The previous devices this type of storage were in experienced no where near the level of reading/writing in standard day-to-day computing tasks, let alone professional work. It may have been fine for an iPod or digital still camera, but people use personal computers data storage in far different ways. Speaking of which, I have a SD card I bought 2 years ago, rarely used, that will occasionally lose a picture, this isn't unheard of with NAND devices (anecdotes ≠ data, but still: devices are not perfect).

In addition, something is not "proven" based on the fact that known issues theoretical or otherwise have been ironed out. With complicated microscopic technology, you don't necessarily know until you build and stress test it over time. Theory gets you only so far--and "proof" is certainly not something you get out of theory.

Even platter hard drives--which have been around much longer--still have major reliability issues with the newest and largest capacities. Are we supposed to believe SSD is magically immune to error?

Yes, theoretically SSD should last much longer. It has no moving parts, of COURSE it'll someday be better--it's common sense.

People should stick to facts and accurate real-world interpretation of them.

Fact: SSD has not been around long enough in real-world capacity to judge its longevity. Most of the reliability information comes from the manufacturers themselves.

Fact: SSD has had serious issues in the past.

Fact: Regardless of what measures have been taken in the manufacturing process, real world data does not yet exist to prove exactly how reliable this technology is. This is especially logical to demand not only because of the history of defective SSDs but also because they're expensive as hell (people with money to burn can experiment all they want).

You're saying things like "real world" and "accurate". Well you're twisting that around to dodge the fact that there's nothing "real world" about SSD reliability data right now. You're taking it on faith from the same companies that sold customers defective hardware in the past. At the time, I'm sure they were saying the same things about theoretical longevity etc etc.

 

[VirtualRain]

Some links to sources that support your view that current gen SSDs are unproven and unworthy of use in desktop systems with testing to back it up would be appreciated.

As far as I can tell, no one has spent more time testing and analyzing SSDs than AnandTech and they are perhaps the biggest advocates of SSDs.

 

[slughead]

Some links to sources that support your view that current gen SSDs are unproven

Well I don't have to tell you what you already know: they haven't been around long enough, and the only valid test is time and the number of units on the market being used by consumers.

Please, show me a few thousand SSD drives that have been used in desktops regularly over the past 4 years. An insanely high failure rate of 10% over 4 years would be extremely difficult to discover given the marketshare and the length of time these devices have been available.

Are you denying any of this? Do I need to supply links that say this is new technology that is currently unpopular? Maybe a few links on statistics? Are none of these things obvious?

This is turning into a religious argument.

 

[TheStrudel]

Anandtech regularly posts that they can't make claims about the long term reliability of these drives for the reasons you've been discussing.

One of the things I like about outfits like Anandtech and Ars Technica (the only really reliable larger computer tech review sites at this point) is that they tell you exactly what they did, qualify every bit of uncertainty, and explain everything.

If it hasn't already been made obvious, Nanofrog's viewpoint is comprehensive, backed up by articles found all over the place, and can be taken at face value.

There is room for debate about the long term reliability of SSDs, and Anandtech have published that as well. More to the point, new SSD technologies are continually being debuted in the marketplace (controllers, firmware), such as the newer Sandforce-based drives, so it's not as simple as saying that SSDs are now proven, because today's SSDs are actually pretty different from what was shipping a few years ago. And they will continue to differentiate themselves.

Unlike hard drives, they are more complicated in every way except mechanical operation and offer far more different approaches and form factors.

Even hard drives are starting to differentiate themselves, slowly, such as recent developments with increasing sector size.

 

[VirtualRain]

Yes. AnandTech is very good in my opinion and they believe Intels claims that an SSD is good for 5+ years at 100GB per day of write activity. This seems consistent with Nanofrog and the vendors rating of 1E4 write cycles since that means a cell can be written to 10000 times before it becomes a read only memory cell. So even if you wiped and filled your drive every day, it would take 28 years to wear it out.

Now As I said, I'm not aware of any source (although I'm open to enlightening myself) that believes this kind of lifespan is unacceptable for desktop use. In fact, it's just the opposite as I said before that credible sources such as Anand who have thouroughly analyzed SSDs seem to recommend them enthusiastically.

 

[JesterJJZ]

Edit: and btw: hybrids are a scam

How? I have a Seagate XT in my Win 7 laptop and it's really snappy now. Big improvement over my old drive.

 

[nanofrog]

Yes. AnandTech is very good in my opinion and they believe Intels claims that an SSD is good for 5+ years at 100GB per day of write activity.

That's manufacturer data though, and not based on real world scenarios (i.e. most of the disk is full with OS and applications). Intel's data was determined on an empty drive. This is where I have a problem, as that's not anything near real world conditions for most usages.

Write amplification occurs at a much higher rate on the remaining capacity (cells not used for OS or application data).

As a result of the increased write amplification on MLC based disks, 3yrs seems more realistic to me based on 60GB filled (past that in terms of writes is possible on remaining capacity not used for applications or OS, but will depend on the specific usage).

Consumer systems won't be used for high writes in most cases (how I think of most consumers' usage patterns), but in an enterprise environment, it wouldn't last that long, and why SLC is needed in such cases (still expected to only last 3yrs under such conditions, and happens to correlate with the typical MTBR implemented for such equipment). Actually, the correlation isn't happenstance IMO, but what the designers were aiming for, as they're well aware that is the usual lenght of time for MTBR.

Without real long-term data on existing generation SSD's, all that's possible are projections.

Now As I said, I'm not aware of any source (although I'm open to enlightening myself) that believes this kind of lifespan is unacceptable for desktop use. In fact, it's just the opposite as I said before that credible sources such as Anand who have thoroughly analyzed SSDs seem to recommend them enthusiastically.

3 - 5yrs is fine for consumer use. But this is also based on certain assumptions (not high write environments on the remaining capacity not consumed by the OS and applications).